On the Design of the LMS Algorithm for Robustness to Outliers in Super-Resolution Video Reconstruction

Super-resolution reconstruction of image sequences is highly dependent on the data outliers and on the quality of the motion estimation. This work addresses the design of the least mean square (LMS) algorithm applied to super-resolution reconstruction. Based on a statistical model of the algorithm behavior, we propose a design strategy to reduce the effects of outliers on the reconstructed image sequence. We show that the proposed strategy can improve the algorithm performance in both transient and steady-state phases of adaptation in practical situations, when registration errors are considered.